Hierarchical Bayesian Model for Product Recommendation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation algorithms either ignore hierarchies, leading to inaccuracies, or use flat hierarchical structures that introduce noise and expand the feature space, failing to effectively capture hierarchical relationships in product recommendations.
Innovation Solution
A hierarchical Bayesian model is defined and trained using a computing device, incorporating features of products, brands, user actions, and user-specific information to predict user actions and provide personalized product recommendations, capturing latent and structural relations through variational Bayes approximation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If flat hierarchical structures are used in recommendation algorithms, then the feature space is expanded and more comprehensive product information is captured, but noise is introduced and model complexity increases leading to reduced recommendation accuracy
Solution Approach 1:
The patent segments the hierarchical product information into multiple levels (category level, brand level, product level) and processes each level separately through dedicated neural network layers. This segmentation allows the model to capture comprehensive features at each hierarchy level while preventing noise propagation across levels, thereby maintaining recommendation accuracy despite expanded feature space coverage.
2Device complexity
If hierarchies are completely ignored in recommendation algorithms, then model complexity is reduced and training is simplified, but recommendation accuracy deteriorates due to loss of hierarchical relationships
Solution Approach 1:
The patent implements a nested hierarchical structure where category embeddings are embedded within brand embeddings, which are in turn embedded within product embeddings. Each level of the hierarchy is nested within the next level, allowing the model to maintain simplified computation at each stage while progressively incorporating hierarchical relationships, thus improving recommendation accuracy without proportionally increasing model complexity.
3Measurement precision
If hierarchical relationships are effectively captured in product recommendations, then recommendation accuracy and precision are improved, but the feature space expands and model complexity increases
Solution Approach 1:
The patent transforms hierarchical relationships from a complex structured problem into a dimensional embedding problem. By representing categories, brands, and products as vectors in a shared embedding space with learned dimensional relationships, the model captures hierarchical patterns through geometric relationships in the embedding space rather than through complex explicit hierarchical structures, thereby improving accuracy while controlling model complexity.
Data Source
AI summary
A method and system for product recommendation. The method includes: defining, by a computing device, a hierarchical Bayesian model having a latent factor; training, by the computing device, the hierarchical Bayesian model using a plurality of training events to obtain a trained hierarchical Bayesian model, each event comprising feature of a product, brand of the product, feature of a user, and action of the user upon the product; predicting, by the computing device, a possibility a target user performing an action on a target product using the trained hierarchical Bayesian model; and providing product recommendation to the target user based on the possibility.